Paper2Agent turns research papers into interactive AI agents

Paper2Agent, an automated framework that converts research papers into interactive AI agents, generated 22 tools for the AlphaGenome model in about 45 minutes at a compute cost of $14, with all tools passing validation, according to a Nature paper published on 16 September 2026.

Five researchers on our who's who tracker had circulated the link by the time we picked it up.

The system, from Jiacheng Miao, Jonathan K. Pritchard, James Zou and co-authors, treats a paper's manuscript, supplementary materials, datasets and code as callable tools exposed through a Model Context Protocol (MCP) server. Once packaged, the resulting agent can be plugged into chat systems such as Claude and queried in natural language.

The authors argue that today's papers are 'fundamentally passive objects', and that Paper2Agent turns them into 'agent-native research objects rather than static documents'. On tutorial-derived queries the AlphaGenome agent scored 98.7 ± 1.3% accuracy and 100.0 ± 0.0% on novel queries; a Scanpy agent for single-cell transcriptomics analysis generated 7 tools in about 45 minutes for $13.

Across a wider test, 74 of 100 computational-biology papers were successfully agentified and 593 of 599 proposed tools passed validation, with 91.2 ± 1.6% accuracy on 300 tutorial-derived questions and a 1.9× runtime speedup over a Claude-plus-repo baseline. Case studies include identifying GPR137 as the causal gene for the psoriasis-associated variant rs887314, ADHD genome-wide association analyses, and interpretation of the LDL cholesterol variant chr1:109274968:G>T.

'Paper agents', the authors conclude, 'could form a dynamic layer of scientific intelligence, accelerating connections across disciplines.'